Convolutional Neural Networks

Project: Write an Algorithm for a Dog Identification App


In this notebook, some template code has already been provided for you, and you will need to implement additional functionality to successfully complete this project. You will not need to modify the included code beyond what is requested. Sections that begin with '(IMPLEMENTATION)' in the header indicate that the following block of code will require additional functionality which you must provide. Instructions will be provided for each section, and the specifics of the implementation are marked in the code block with a 'TODO' statement. Please be sure to read the instructions carefully!

Note: Once you have completed all of the code implementations, you need to finalize your work by exporting the iPython Notebook as an HTML document. Before exporting the notebook to html, all of the code cells need to have been run so that reviewers can see the final implementation and output. You can then export the notebook by using the menu above and navigating to \n", "File -> Download as -> HTML (.html). Include the finished document along with this notebook as your submission.

In addition to implementing code, there will be questions that you must answer which relate to the project and your implementation. Each section where you will answer a question is preceded by a 'Question X' header. Carefully read each question and provide thorough answers in the following text boxes that begin with 'Answer:'. Your project submission will be evaluated based on your answers to each of the questions and the implementation you provide.

Note: Code and Markdown cells can be executed using the Shift + Enter keyboard shortcut. Markdown cells can be edited by double-clicking the cell to enter edit mode.

The rubric contains optional "Stand Out Suggestions" for enhancing the project beyond the minimum requirements. If you decide to pursue the "Stand Out Suggestions", you should include the code in this IPython notebook.


Why We're Here

In this notebook, you will make the first steps towards developing an algorithm that could be used as part of a mobile or web app. At the end of this project, your code will accept any user-supplied image as input. If a dog is detected in the image, it will provide an estimate of the dog's breed. If a human is detected, it will provide an estimate of the dog breed that is most resembling. The image below displays potential sample output of your finished project (... but we expect that each student's algorithm will behave differently!).

Sample Dog Output

In this real-world setting, you will need to piece together a series of models to perform different tasks; for instance, the algorithm that detects humans in an image will be different from the CNN that infers dog breed. There are many points of possible failure, and no perfect algorithm exists. Your imperfect solution will nonetheless create a fun user experience!

The Road Ahead

We break the notebook into separate steps. Feel free to use the links below to navigate the notebook.

  • Step 0: Import Datasets
  • Step 1: Detect Humans
  • Step 2: Detect Dogs
  • Step 3: Create a CNN to Classify Dog Breeds (from Scratch)
  • Step 4: Use a CNN to Classify Dog Breeds (using Transfer Learning)
  • Step 5: Create a CNN to Classify Dog Breeds (using Transfer Learning)
  • Step 6: Write your Algorithm
  • Step 7: Test Your Algorithm

Step 0: Import Datasets

Import Dog Dataset

In the code cell below, we import a dataset of dog images. We populate a few variables through the use of the load_files function from the scikit-learn library:

  • train_files, valid_files, test_files - numpy arrays containing file paths to images
  • train_targets, valid_targets, test_targets - numpy arrays containing onehot-encoded classification labels
  • dog_names - list of string-valued dog breed names for translating labels
In [1]:
from sklearn.datasets import load_files       
from keras.utils import np_utils
import numpy as np
from glob import glob
import os

cwd = os.getcwd()

# define function to load train, test, and validation datasets
def load_dataset(path):
    path = os.path.join(cwd, path)
    data = load_files(path)
    dog_files = np.array(data['filenames'])
    dog_targets = np_utils.to_categorical(np.array(data['target']), 133)
    return dog_files, dog_targets

# load train, test, and validation datasets
train_files, train_targets = load_dataset('data/dog_images/train')
valid_files, valid_targets = load_dataset('data/dog_images/valid')
test_files, test_targets = load_dataset('data/dog_images/test')

# load list of dog names
dog_names = [item[20:-1] for item in sorted(glob("data/dog_images/train/*/"))]

# print statistics about the dataset
print('There are %d total dog categories.' % len(dog_names))
print('There are %s total dog images.\n' % len(np.hstack([train_files, valid_files, test_files])))
print('There are %d training dog images.' % len(train_files))
print('There are %d validation dog images.' % len(valid_files))
print('There are %d test dog images.'% len(test_files))
Using TensorFlow backend.
There are 133 total dog categories.
There are 8351 total dog images.

There are 6680 training dog images.
There are 835 validation dog images.
There are 836 test dog images.

Import Human Dataset

In the code cell below, we import a dataset of human images, where the file paths are stored in the numpy array human_files.

In [2]:
import random
random.seed(8675309)

# load filenames in shuffled human dataset
human_files = np.array(glob("data/lfw/*/*"))
random.shuffle(human_files)

# print statistics about the dataset
print('There are %d total human images.' % len(human_files))
There are 13233 total human images.

Step 1: Detect Humans

We use OpenCV's implementation of Haar feature-based cascade classifiers to detect human faces in images. OpenCV provides many pre-trained face detectors, stored as XML files on github. We have downloaded one of these detectors and stored it in the haarcascades directory.

In the next code cell, we demonstrate how to use this detector to find human faces in a sample image.

In [3]:
import cv2                
import matplotlib.pyplot as plt                        
%matplotlib inline                               

# extract pre-trained face detector
face_cascade = cv2.CascadeClassifier('haarcascades/haarcascade_frontalface_alt.xml')

# load color (BGR) image
img = cv2.imread(human_files[3])
# convert BGR image to grayscale
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

# find faces in image
faces = face_cascade.detectMultiScale(gray)

# print number of faces detected in the image
print('Number of faces detected:', len(faces))

# get bounding box for each detected face
for (x,y,w,h) in faces:
    # add bounding box to color image
    cv2.rectangle(img,(x,y),(x+w,y+h),(255,0,0),2)
    
# convert BGR image to RGB for plotting
cv_rgb = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)

# display the image, along with bounding box
plt.imshow(cv_rgb)
plt.show()
Number of faces detected: 1

Before using any of the face detectors, it is standard procedure to convert the images to grayscale. The detectMultiScale function executes the classifier stored in face_cascade and takes the grayscale image as a parameter.

In the above code, faces is a numpy array of detected faces, where each row corresponds to a detected face. Each detected face is a 1D array with four entries that specifies the bounding box of the detected face. The first two entries in the array (extracted in the above code as x and y) specify the horizontal and vertical positions of the top left corner of the bounding box. The last two entries in the array (extracted here as w and h) specify the width and height of the box.

Write a Human Face Detector

We can use this procedure to write a function that returns True if a human face is detected in an image and False otherwise. This function, aptly named face_detector, takes a string-valued file path to an image as input and appears in the code block below.

In [4]:
# returns "True" if face is detected in image stored at img_path
def face_detector(img_path):
    img = cv2.imread(img_path)
    gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
    faces = face_cascade.detectMultiScale(gray)
    return len(faces) > 0

(IMPLEMENTATION) Assess the Human Face Detector

Question 1: Use the code cell below to test the performance of the face_detector function.

  • What percentage of the first 100 images in human_files have a detected human face?
  • What percentage of the first 100 images in dog_files have a detected human face?

Ideally, we would like 100% of human images with a detected face and 0% of dog images with a detected face. You will see that our algorithm falls short of this goal, but still gives acceptable performance. We extract the file paths for the first 100 images from each of the datasets and store them in the numpy arrays human_files_short and dog_files_short.

Answer: The face_detector function detects human face in 100% of the first 100 human_files and in 11% of the first 100 dog_files.

In [5]:
human_files_short = human_files[:100]
dog_files_short = train_files[:100]
# Do NOT modify the code above this line.

## TODO: Test the performance of the face_detector algorithm 
## on the images in human_files_short and dog_files_short.
def human_face_perc(files_list):
    count = 0
    for f in files_list:
        if face_detector(f):
            count += 1
    return float(count) * 100.0 / len(files_list)

print ('Percentage of human faces in human_files: {:.2f}'.format(human_face_perc(human_files_short)),
      '\nPercentage of human faces in dog_files: {:.2f}'.format(human_face_perc(dog_files_short)))
Percentage of human faces in human_files: 99.00 
Percentage of human faces in dog_files: 12.00

Question 2: This algorithmic choice necessitates that we communicate to the user that we accept human images only when they provide a clear view of a face (otherwise, we risk having unneccessarily frustrated users!). In your opinion, is this a reasonable expectation to pose on the user? If not, can you think of a way to detect humans in images that does not necessitate an image with a clearly presented face?

Answer: I think the answer here depends on the requirement. If we cannot see any feature in the image, then we cannot form an opinion about the feature. For example, in case of a side view, are there any marks on the other side of the face, or the colour of the other eye? If we need to detect the human image for documentation purposes such as passport, then we should expect a clear veiw of a face. However, if we need this to tag people in pictures on say, facebook, then this shouldn't be a requirement. There are situations when we would like to be somewhere in the middle. For example, the face recognition technology on the latest iPhones require a substantial amount of face being visible. If you are out in the cold wearing a cap or in the sun wearing glasses, you want the phone to still detect you. But the phone shouldn't detect you if you are not looking at the phone or sleeping.

In order to detect humans without a clear presented face, we could train a neural network by presenting images of people with a partially hidden face. One way to do this is by using a full visible face but some of the nodes in the network, specially at the input or initial layers are dropped out. This would represent a less than clear presentation of face and we could try to train our network with such drop outs.


The face detector from OpenCV as a potential way to detect human images will be used further in our algorithm


Step 2: Detect Dogs

In this section, we use a pre-trained ResNet-50 model to detect dogs in images. Our first line of code downloads the ResNet-50 model, along with weights that have been trained on ImageNet, a very large, very popular dataset used for image classification and other vision tasks. ImageNet contains over 10 million URLs, each linking to an image containing an object from one of 1000 categories. Given an image, this pre-trained ResNet-50 model returns a prediction (derived from the available categories in ImageNet) for the object that is contained in the image.

In [6]:
from keras.applications.resnet50 import ResNet50

# define ResNet50 model
ResNet50_model = ResNet50(weights='imagenet')

Pre-process the Data

When using TensorFlow as backend, Keras CNNs require a 4D array (which we'll also refer to as a 4D tensor) as input, with shape

$$ (\text{nb_samples}, \text{rows}, \text{columns}, \text{channels}), $$

where nb_samples corresponds to the total number of images (or samples), and rows, columns, and channels correspond to the number of rows, columns, and channels for each image, respectively.

The path_to_tensor function below takes a string-valued file path to a color image as input and returns a 4D tensor suitable for supplying to a Keras CNN. The function first loads the image and resizes it to a square image that is $224 \times 224$ pixels. Next, the image is converted to an array, which is then resized to a 4D tensor. In this case, since we are working with color images, each image has three channels. Likewise, since we are processing a single image (or sample), the returned tensor will always have shape

$$ (1, 224, 224, 3). $$

The paths_to_tensor function takes a numpy array of string-valued image paths as input and returns a 4D tensor with shape

$$ (\text{nb_samples}, 224, 224, 3). $$

Here, nb_samples is the number of samples, or number of images, in the supplied array of image paths. It is best to think of nb_samples as the number of 3D tensors (where each 3D tensor corresponds to a different image) in your dataset!

In [7]:
from keras.preprocessing import image                  
from tqdm import tqdm

def path_to_tensor(img_path):
    # loads RGB image as PIL.Image.Image type
    img = image.load_img(img_path, target_size=(224, 224))
    # convert PIL.Image.Image type to 3D tensor with shape (224, 224, 3)
    x = image.img_to_array(img)
    # convert 3D tensor to 4D tensor with shape (1, 224, 224, 3) and return 4D tensor
    return np.expand_dims(x, axis=0)

def paths_to_tensor(img_paths):
    list_of_tensors = [path_to_tensor(img_path) for img_path in tqdm(img_paths)]
    return np.vstack(list_of_tensors)

Making Predictions with ResNet-50

Getting the 4D tensor ready for ResNet-50, and for any other pre-trained model in Keras, requires some additional processing. First, the RGB image is converted to BGR by reordering the channels. All pre-trained models have the additional normalization step that the mean pixel (expressed in RGB as $[103.939, 116.779, 123.68]$ and calculated from all pixels in all images in ImageNet) must be subtracted from every pixel in each image. This is implemented in the imported function preprocess_input. If you're curious, you can check the code for preprocess_input here.

Now that we have a way to format our image for supplying to ResNet-50, we are now ready to use the model to extract the predictions. This is accomplished with the predict method, which returns an array whose $i$-th entry is the model's predicted probability that the image belongs to the $i$-th ImageNet category. This is implemented in the ResNet50_predict_labels function below.

By taking the argmax of the predicted probability vector, we obtain an integer corresponding to the model's predicted object class, which we can identify with an object category through the use of this dictionary.

In [8]:
from keras.applications.resnet50 import preprocess_input, decode_predictions

def ResNet50_predict_labels(img_path):
    # returns prediction vector for image located at img_path
    img = preprocess_input(path_to_tensor(img_path))
    return np.argmax(ResNet50_model.predict(img))

Write a Dog Detector

While looking at the dictionary, you will notice that the categories corresponding to dogs appear in an uninterrupted sequence and correspond to dictionary keys 151-268, inclusive, to include all categories from 'Chihuahua' to 'Mexican hairless'. Thus, in order to check to see if an image is predicted to contain a dog by the pre-trained ResNet-50 model, we need only check if the ResNet50_predict_labels function above returns a value between 151 and 268 (inclusive).

We use these ideas to complete the dog_detector function below, which returns True if a dog is detected in an image (and False if not).

In [9]:
### returns "True" if a dog is detected in the image stored at img_path
def dog_detector(img_path):
    prediction = ResNet50_predict_labels(img_path)
    return ((prediction <= 268) & (prediction >= 151)) 

(IMPLEMENTATION) Assess the Dog Detector

Question 3: Use the code cell below to test the performance of your dog_detector function.

  • What percentage of the images in human_files_short have a detected dog?
  • What percentage of the images in dog_files_short have a detected dog?

Answer: The dog_detector function detects dog face in 0% (None) of the first 100 human_files and in 100% of the first 100 dog_files.

In [10]:
### TODO: Test the performance of the dog_detector function
### on the images in human_files_short and dog_files_short.

def dog_perc(files_list):
    count = 0
    for f in files_list:
        if dog_detector(f):
            count += 1
    return float(count) * 100.0 / len(files_list)

print ('Percentage of dog faces in human_files_short: {:.2f}'.format(dog_perc(human_files_short)),
      '\nPercentage of dog faces in dog_files_short: {:.2f}'.format(dog_perc(dog_files_short)))
Percentage of dog faces in human_files_short: 1.00 
Percentage of dog faces in dog_files_short: 100.00

Step 3: Create a CNN to Classify Dog Breeds (from Scratch)

Now that we have functions for detecting humans and dogs in images, we need a way to predict breed from images. In this step, you will create a CNN that classifies dog breeds. You must create your CNN from scratch (so, you can't use transfer learning yet!), and you must attain a test accuracy of at least 1%. In Step 5 of this notebook, you will have the opportunity to use transfer learning to create a CNN that attains greatly improved accuracy.

Be careful with adding too many trainable layers! More parameters means longer training, which means you are more likely to need a GPU to accelerate the training process. Thankfully, Keras provides a handy estimate of the time that each epoch is likely to take; you can extrapolate this estimate to figure out how long it will take for your algorithm to train.

We mention that the task of assigning breed to dogs from images is considered exceptionally challenging. To see why, consider that even a human would have great difficulty in distinguishing between a Brittany and a Welsh Springer Spaniel.

Brittany Welsh Springer Spaniel

It is not difficult to find other dog breed pairs with minimal inter-class variation (for instance, Curly-Coated Retrievers and American Water Spaniels).

Curly-Coated Retriever American Water Spaniel

Likewise, recall that labradors come in yellow, chocolate, and black. Your vision-based algorithm will have to conquer this high intra-class variation to determine how to classify all of these different shades as the same breed.

Yellow Labrador Chocolate Labrador Black Labrador

We also mention that random chance presents an exceptionally low bar: setting aside the fact that the classes are slightly imabalanced, a random guess will provide a correct answer roughly 1 in 133 times, which corresponds to an accuracy of less than 1%.

Remember that the practice is far ahead of the theory in deep learning. Experiment with many different architectures, and trust your intuition. And, of course, have fun!

Pre-process the Data

We rescale the images by dividing every pixel in every image by 255.

In [11]:
from PIL import ImageFile                            
ImageFile.LOAD_TRUNCATED_IMAGES = True                 

# pre-process the data for Keras
train_tensors = paths_to_tensor(train_files).astype('float32')/255
valid_tensors = paths_to_tensor(valid_files).astype('float32')/255
test_tensors = paths_to_tensor(test_files).astype('float32')/255
100%|█████████████████████████████████████████████████████████████████████████████| 6680/6680 [00:36<00:00, 184.37it/s]
100%|███████████████████████████████████████████████████████████████████████████████| 835/835 [00:04<00:00, 201.97it/s]
100%|███████████████████████████████████████████████████████████████████████████████| 836/836 [00:04<00:00, 203.77it/s]

(IMPLEMENTATION) Model Architecture

Create a CNN to classify dog breed. At the end of your code cell block, summarize the layers of your model by executing the line:

    model.summary()

We have imported some Python modules to get you started, but feel free to import as many modules as you need. If you end up getting stuck, here's a hint that specifies a model that trains relatively fast on CPU and attains >1% test accuracy in 5 epochs:

Sample CNN

Question 4: Outline the steps you took to get to your final CNN architecture and your reasoning at each step. If you chose to use the hinted architecture above, describe why you think that CNN architecture should work well for the image classification task.

Answer: I started with 3 convolutional layers (1 Conv2D + 1 MaxPooling2D + 1 Dropout) as the initial 2 layers capture simple features and the 3rd layer will capture the more complex features. The first layer has 16 filters, second has 32 and third has 64 filters, so essentially doubling up the filters each time. I also reduce the height and width by half with each layer using the maxpooling2d layer (2x2 pool size and stride of 2) and further use a Dropout layer with a rate of 0.2. Then I use a flatten layer (layer 4), followed by a dense layer of 500 nodes (layer 6) and a final layer of 133 nodes (layer 7). I chose 500 nodes for layer 6 so that it is much lower than the nodes in flatten layer but reasonably bigger than the number of output nodes. All layers have relu filter, except for the output whic has the softmax filter.

After running the summary, I realised that the number of parameters is over 25 million, which is very high. This number is dominated by the dense layer (layer 6). To reduce this, I add another convolutional layer with 64 filters (keeping same as earlier layer) and reducing the height and width by 2 using the maxpooling2d layer. I add a dropout layer with rate of 0.2 as well. The additional layer should also assist in identifying complex features. Keeping rest same, I get 6 million parameters which still looks a bit high but I wish to see how it performance.

After training, I am able to achieve more than 5.5% accuracy on the test set. I used 40 epochs with the ability to early stop if the validation loss doesn't improve for 10 consecutive epochs. I have also plotted the training and testing (validation) accuracy against the epochs to see when the overfitting happens. I have also plotted this for the training and testing (validation) loss. The model overfits after 4 epochs - I can see that the training loss keeps reducing but the validation loss increases after 4 epochs.

In [12]:
from keras.layers import Conv2D, MaxPooling2D, GlobalAveragePooling2D
from keras.layers import Dropout, Flatten, Dense
from keras.models import Sequential

model = Sequential()

### TODO: Define your architecture.
# Layer 1
model.add(Conv2D(filters = 16, kernel_size = 3, padding = 'same', activation = 'relu', input_shape=(224,224,3)))
model.add(MaxPooling2D(pool_size = 2, strides = 2))
model.add(Dropout(0.2))
# Layer 2
model.add(Conv2D(filters = 32, kernel_size = 3, padding = 'same', activation = 'relu'))
model.add(MaxPooling2D(pool_size = 2, strides = 2))
model.add(Dropout(0.2))
# Layer 3
model.add(Conv2D(filters = 64, kernel_size = 3, padding = 'same', activation = 'relu'))
model.add(MaxPooling2D(pool_size = 2, strides = 2))
model.add(Dropout(0.2))
# Layer 4
model.add(Conv2D(filters = 64, kernel_size = 3, padding = 'same', activation = 'relu'))
model.add(MaxPooling2D(pool_size = 2, strides = 2))
model.add(Dropout(0.2))
# Layer 5
model.add(Flatten())
# Layer 6
model.add(Dense(500, activation = 'relu'))
# Layer 7
model.add(Dense(133, activation = 'softmax'))
model.summary()
Model: "sequential_1"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
conv2d_1 (Conv2D)            (None, 224, 224, 16)      448       
_________________________________________________________________
max_pooling2d_2 (MaxPooling2 (None, 112, 112, 16)      0         
_________________________________________________________________
dropout_1 (Dropout)          (None, 112, 112, 16)      0         
_________________________________________________________________
conv2d_2 (Conv2D)            (None, 112, 112, 32)      4640      
_________________________________________________________________
max_pooling2d_3 (MaxPooling2 (None, 56, 56, 32)        0         
_________________________________________________________________
dropout_2 (Dropout)          (None, 56, 56, 32)        0         
_________________________________________________________________
conv2d_3 (Conv2D)            (None, 56, 56, 64)        18496     
_________________________________________________________________
max_pooling2d_4 (MaxPooling2 (None, 28, 28, 64)        0         
_________________________________________________________________
dropout_3 (Dropout)          (None, 28, 28, 64)        0         
_________________________________________________________________
conv2d_4 (Conv2D)            (None, 28, 28, 64)        36928     
_________________________________________________________________
max_pooling2d_5 (MaxPooling2 (None, 14, 14, 64)        0         
_________________________________________________________________
dropout_4 (Dropout)          (None, 14, 14, 64)        0         
_________________________________________________________________
flatten_1 (Flatten)          (None, 12544)             0         
_________________________________________________________________
dense_1 (Dense)              (None, 500)               6272500   
_________________________________________________________________
dense_2 (Dense)              (None, 133)               66633     
=================================================================
Total params: 6,399,645
Trainable params: 6,399,645
Non-trainable params: 0
_________________________________________________________________

Compile the Model

In [13]:
model.compile(optimizer='adam', loss='categorical_crossentropy', metrics=['accuracy'])

(IMPLEMENTATION) Train the Model

Train your model in the code cell below. Use model checkpointing to save the model that attains the best validation loss.

You are welcome to augment the training data, but this is not a requirement.

In [15]:
from keras.callbacks import ModelCheckpoint, EarlyStopping  

### TODO: specify the number of epochs that you would like to use to train the model.

epochs = 40

### Do NOT modify the code below this line.

checkpointer_1 = ModelCheckpoint(filepath='saved_models/weights.best.from_scratch.hdf5', monitor = 'val_loss', verbose=1, 
                               save_best_only=True)

checkpointer_2 = EarlyStopping(monitor = 'val_loss', verbose = 1, patience = 10)

calcs = model.fit(train_tensors, train_targets, 
          validation_data=(valid_tensors, valid_targets),
          epochs=epochs, batch_size=20, callbacks=[checkpointer_1, checkpointer_2], verbose=1)

# summarize history for accuracy
plt.plot(calcs.history['accuracy'])
plt.plot(calcs.history['val_accuracy'])
plt.title('Model Accuracy')
plt.ylabel('Accuracy')
plt.xlabel('epoch')
plt.legend(['train', 'test'], loc='upper left')
plt.show()

# summarize history for loss
plt.plot(calcs.history['loss'])
plt.plot(calcs.history['val_loss'])
plt.title('Model Loss')
plt.ylabel('Loss')
plt.xlabel('epoch')
plt.legend(['train', 'test'], loc='upper left')
plt.show()
Train on 6680 samples, validate on 835 samples
Epoch 1/40
6680/6680 [==============================] - 9s 1ms/step - loss: 0.1367 - accuracy: 0.9611 - val_loss: 10.0575 - val_accuracy: 0.0862

Epoch 00001: val_loss improved from inf to 10.05752, saving model to saved_models/weights.best.from_scratch.hdf5
Epoch 2/40
6680/6680 [==============================] - 9s 1ms/step - loss: 0.1696 - accuracy: 0.9587 - val_loss: 9.3195 - val_accuracy: 0.0850

Epoch 00002: val_loss improved from 10.05752 to 9.31949, saving model to saved_models/weights.best.from_scratch.hdf5
Epoch 3/40
6680/6680 [==============================] - 9s 1ms/step - loss: 0.1144 - accuracy: 0.9710 - val_loss: 9.7522 - val_accuracy: 0.0994

Epoch 00003: val_loss did not improve from 9.31949
Epoch 4/40
6680/6680 [==============================] - 9s 1ms/step - loss: 0.1281 - accuracy: 0.9606 - val_loss: 10.1183 - val_accuracy: 0.0970

Epoch 00004: val_loss did not improve from 9.31949
Epoch 5/40
6680/6680 [==============================] - 9s 1ms/step - loss: 0.0985 - accuracy: 0.9713 - val_loss: 10.2074 - val_accuracy: 0.0886

Epoch 00005: val_loss did not improve from 9.31949
Epoch 6/40
6680/6680 [==============================] - 9s 1ms/step - loss: 0.0873 - accuracy: 0.9753 - val_loss: 10.3833 - val_accuracy: 0.1006

Epoch 00006: val_loss did not improve from 9.31949
Epoch 7/40
6680/6680 [==============================] - 9s 1ms/step - loss: 0.0969 - accuracy: 0.9731 - val_loss: 9.5506 - val_accuracy: 0.0850

Epoch 00007: val_loss did not improve from 9.31949
Epoch 8/40
6680/6680 [==============================] - 9s 1ms/step - loss: 0.1129 - accuracy: 0.9680 - val_loss: 9.1509 - val_accuracy: 0.0862

Epoch 00008: val_loss improved from 9.31949 to 9.15086, saving model to saved_models/weights.best.from_scratch.hdf5
Epoch 9/40
6680/6680 [==============================] - 9s 1ms/step - loss: 0.1010 - accuracy: 0.9723 - val_loss: 9.2314 - val_accuracy: 0.0970

Epoch 00009: val_loss did not improve from 9.15086
Epoch 10/40
6680/6680 [==============================] - 9s 1ms/step - loss: 0.0758 - accuracy: 0.9753 - val_loss: 10.9134 - val_accuracy: 0.0743

Epoch 00010: val_loss did not improve from 9.15086
Epoch 11/40
6680/6680 [==============================] - 9s 1ms/step - loss: 0.1228 - accuracy: 0.9659 - val_loss: 9.9221 - val_accuracy: 0.0886

Epoch 00011: val_loss did not improve from 9.15086
Epoch 12/40
6680/6680 [==============================] - 9s 1ms/step - loss: 0.0909 - accuracy: 0.9772 - val_loss: 10.3709 - val_accuracy: 0.0958

Epoch 00012: val_loss did not improve from 9.15086
Epoch 13/40
6680/6680 [==============================] - 9s 1ms/step - loss: 0.0871 - accuracy: 0.9786 - val_loss: 9.9038 - val_accuracy: 0.0766

Epoch 00013: val_loss did not improve from 9.15086
Epoch 14/40
6680/6680 [==============================] - 9s 1ms/step - loss: 0.0717 - accuracy: 0.9799 - val_loss: 10.3888 - val_accuracy: 0.0946

Epoch 00014: val_loss did not improve from 9.15086
Epoch 15/40
6680/6680 [==============================] - 9s 1ms/step - loss: 0.0977 - accuracy: 0.9702 - val_loss: 10.7550 - val_accuracy: 0.0731

Epoch 00015: val_loss did not improve from 9.15086
Epoch 16/40
6680/6680 [==============================] - 9s 1ms/step - loss: 0.0772 - accuracy: 0.9766 - val_loss: 10.3006 - val_accuracy: 0.0886

Epoch 00016: val_loss did not improve from 9.15086
Epoch 17/40
6680/6680 [==============================] - 9s 1ms/step - loss: 0.0748 - accuracy: 0.9778 - val_loss: 10.6872 - val_accuracy: 0.0826

Epoch 00017: val_loss did not improve from 9.15086
Epoch 18/40
6680/6680 [==============================] - 9s 1ms/step - loss: 0.0680 - accuracy: 0.9808 - val_loss: 10.6938 - val_accuracy: 0.0898

Epoch 00018: val_loss did not improve from 9.15086
Epoch 00018: early stopping

Load the Model with the Best Validation Loss

In [16]:
model.load_weights('saved_models/weights.best.from_scratch.hdf5')

Test the Model

Try out your model on the test dataset of dog images. Ensure that your test accuracy is greater than 1%.

In [17]:
# get index of predicted dog breed for each image in test set
dog_breed_predictions = [np.argmax(model.predict(np.expand_dims(tensor, axis=0))) for tensor in test_tensors]

# report test accuracy
test_accuracy = 100*np.sum(np.array(dog_breed_predictions)==np.argmax(test_targets, axis=1))/len(dog_breed_predictions)
print('Test accuracy: %.4f%%' % test_accuracy)
Test accuracy: 8.7321%

Step 4: Use a CNN to Classify Dog Breeds

To reduce training time without sacrificing accuracy, we show you how to train a CNN using transfer learning. In the following step, you will get a chance to use transfer learning to train your own CNN.

Obtain Bottleneck Features

In [18]:
bottleneck_features = np.load('data/bottleneck_features/DogVGG16Data.npz')
train_VGG16 = bottleneck_features['train']
valid_VGG16 = bottleneck_features['valid']
test_VGG16 = bottleneck_features['test']

Model Architecture

The model uses the the pre-trained VGG-16 model as a fixed feature extractor, where the last convolutional output of VGG-16 is fed as input to our model. We only add a global average pooling layer and a fully connected layer, where the latter contains one node for each dog category and is equipped with a softmax.

In [19]:
VGG16_model = Sequential()
VGG16_model.add(GlobalAveragePooling2D(input_shape=train_VGG16.shape[1:]))
VGG16_model.add(Dense(133, activation='softmax'))

VGG16_model.summary()
Model: "sequential_2"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
global_average_pooling2d_1 ( (None, 512)               0         
_________________________________________________________________
dense_3 (Dense)              (None, 133)               68229     
=================================================================
Total params: 68,229
Trainable params: 68,229
Non-trainable params: 0
_________________________________________________________________

Compile the Model

In [20]:
VGG16_model.compile(loss='categorical_crossentropy', optimizer='rmsprop', metrics=['accuracy'])

Train the Model

In [21]:
checkpointer = ModelCheckpoint(filepath='saved_models/weights.best.VGG16.hdf5', 
                               verbose=1, save_best_only=True)

VGG16_model.fit(train_VGG16, train_targets, 
          validation_data=(valid_VGG16, valid_targets),
          epochs=20, batch_size=20, callbacks=[checkpointer], verbose=1)
Train on 6680 samples, validate on 835 samples
Epoch 1/20
6680/6680 [==============================] - 1s 157us/step - loss: 8.0558 - accuracy: 0.2280 - val_loss: 3.6446 - val_accuracy: 0.4347

Epoch 00001: val_loss improved from inf to 3.64462, saving model to saved_models/weights.best.VGG16.hdf5
Epoch 2/20
6680/6680 [==============================] - 1s 148us/step - loss: 2.2214 - accuracy: 0.5913 - val_loss: 2.4381 - val_accuracy: 0.5952

Epoch 00002: val_loss improved from 3.64462 to 2.43813, saving model to saved_models/weights.best.VGG16.hdf5
Epoch 3/20
6680/6680 [==============================] - 1s 144us/step - loss: 1.2826 - accuracy: 0.7313 - val_loss: 1.9549 - val_accuracy: 0.6515

Epoch 00003: val_loss improved from 2.43813 to 1.95487, saving model to saved_models/weights.best.VGG16.hdf5
Epoch 4/20
6680/6680 [==============================] - 1s 148us/step - loss: 0.8222 - accuracy: 0.8085 - val_loss: 2.1287 - val_accuracy: 0.6611

Epoch 00004: val_loss did not improve from 1.95487
Epoch 5/20
6680/6680 [==============================] - 1s 148us/step - loss: 0.5734 - accuracy: 0.8542 - val_loss: 1.9365 - val_accuracy: 0.6922

Epoch 00005: val_loss improved from 1.95487 to 1.93651, saving model to saved_models/weights.best.VGG16.hdf5
Epoch 6/20
6680/6680 [==============================] - 1s 151us/step - loss: 0.4190 - accuracy: 0.8949 - val_loss: 1.8264 - val_accuracy: 0.7126

Epoch 00006: val_loss improved from 1.93651 to 1.82642, saving model to saved_models/weights.best.VGG16.hdf5
Epoch 7/20
6680/6680 [==============================] - 1s 148us/step - loss: 0.2990 - accuracy: 0.9190 - val_loss: 1.7909 - val_accuracy: 0.7102

Epoch 00007: val_loss improved from 1.82642 to 1.79087, saving model to saved_models/weights.best.VGG16.hdf5
Epoch 8/20
6680/6680 [==============================] - 1s 148us/step - loss: 0.2264 - accuracy: 0.9347 - val_loss: 1.9218 - val_accuracy: 0.7138

Epoch 00008: val_loss did not improve from 1.79087
Epoch 9/20
6680/6680 [==============================] - 1s 147us/step - loss: 0.1812 - accuracy: 0.9507 - val_loss: 1.9215 - val_accuracy: 0.7150

Epoch 00009: val_loss did not improve from 1.79087
Epoch 10/20
6680/6680 [==============================] - 1s 146us/step - loss: 0.1365 - accuracy: 0.9606 - val_loss: 1.9847 - val_accuracy: 0.7234

Epoch 00010: val_loss did not improve from 1.79087
Epoch 11/20
6680/6680 [==============================] - 1s 148us/step - loss: 0.0975 - accuracy: 0.9699 - val_loss: 1.7582 - val_accuracy: 0.7293

Epoch 00011: val_loss improved from 1.79087 to 1.75817, saving model to saved_models/weights.best.VGG16.hdf5
Epoch 12/20
6680/6680 [==============================] - 1s 150us/step - loss: 0.0751 - accuracy: 0.9741 - val_loss: 1.8784 - val_accuracy: 0.7341

Epoch 00012: val_loss did not improve from 1.75817
Epoch 13/20
6680/6680 [==============================] - 1s 152us/step - loss: 0.0604 - accuracy: 0.9826 - val_loss: 1.9897 - val_accuracy: 0.7210

Epoch 00013: val_loss did not improve from 1.75817
Epoch 14/20
6680/6680 [==============================] - 1s 145us/step - loss: 0.0495 - accuracy: 0.9840 - val_loss: 1.9475 - val_accuracy: 0.7234

Epoch 00014: val_loss did not improve from 1.75817
Epoch 15/20
6680/6680 [==============================] - 1s 147us/step - loss: 0.0390 - accuracy: 0.9867 - val_loss: 1.9347 - val_accuracy: 0.7449

Epoch 00015: val_loss did not improve from 1.75817
Epoch 16/20
6680/6680 [==============================] - 1s 148us/step - loss: 0.0401 - accuracy: 0.9883 - val_loss: 1.9935 - val_accuracy: 0.7281

Epoch 00016: val_loss did not improve from 1.75817
Epoch 17/20
6680/6680 [==============================] - 1s 149us/step - loss: 0.0335 - accuracy: 0.9909 - val_loss: 1.9216 - val_accuracy: 0.7353

Epoch 00017: val_loss did not improve from 1.75817
Epoch 18/20
6680/6680 [==============================] - 1s 150us/step - loss: 0.0251 - accuracy: 0.9931 - val_loss: 1.9890 - val_accuracy: 0.7389

Epoch 00018: val_loss did not improve from 1.75817
Epoch 19/20
6680/6680 [==============================] - 1s 147us/step - loss: 0.0250 - accuracy: 0.9933 - val_loss: 2.0620 - val_accuracy: 0.7413

Epoch 00019: val_loss did not improve from 1.75817
Epoch 20/20
6680/6680 [==============================] - 1s 147us/step - loss: 0.0208 - accuracy: 0.9942 - val_loss: 2.1312 - val_accuracy: 0.7413

Epoch 00020: val_loss did not improve from 1.75817
Out[21]:
<keras.callbacks.callbacks.History at 0x1903e1dce88>

Load the Model with the Best Validation Loss

In [22]:
VGG16_model.load_weights('saved_models/weights.best.VGG16.hdf5')

Test the Model

Now, we can use the CNN to test how well it identifies breed within our test dataset of dog images. We print the test accuracy below.

In [23]:
# get index of predicted dog breed for each image in test set
VGG16_predictions = [np.argmax(VGG16_model.predict(np.expand_dims(feature, axis=0))) for feature in test_VGG16]

# report test accuracy
test_accuracy = 100*np.sum(np.array(VGG16_predictions)==np.argmax(test_targets, axis=1))/len(VGG16_predictions)
print('Test accuracy: %.4f%%' % test_accuracy)
Test accuracy: 74.0431%

Predict Dog Breed with the Model

In [24]:
from extract_bottleneck_features import *

def VGG16_predict_breed(img_path):
    # extract bottleneck features
    bottleneck_feature = extract_VGG16(path_to_tensor(img_path))
    # obtain predicted vector
    predicted_vector = VGG16_model.predict(bottleneck_feature)
    # return dog breed that is predicted by the model
    return dog_names[np.argmax(predicted_vector)]

Step 5: Create a CNN to Classify Dog Breeds (using Transfer Learning)

You will now use transfer learning to create a CNN that can identify dog breed from images. Your CNN must attain at least 60% accuracy on the test set.

In Step 4, we used transfer learning to create a CNN using VGG-16 bottleneck features. In this section, you must use the bottleneck features from a different pre-trained model. To make things easier for you, we have pre-computed the features for all of the networks that are currently available in Keras. These are already in the workspace, at /data/bottleneck_features. If you wish to download them on a different machine, they can be found at:

The files are encoded as such:

Dog{network}Data.npz

where {network}, in the above filename, can be one of VGG19, Resnet50, InceptionV3, or Xception.

The above architectures are downloaded and stored for you in the /data/bottleneck_features/ folder.

This means the following will be in the /data/bottleneck_features/ folder:

DogVGG19Data.npz DogResnet50Data.npz DogInceptionV3Data.npz DogXceptionData.npz

(IMPLEMENTATION) Obtain Bottleneck Features

In the code block below, extract the bottleneck features corresponding to the train, test, and validation sets by running the following:

bottleneck_features = np.load('/data/bottleneck_features/Dog{network}Data.npz')
train_{network} = bottleneck_features['train']
valid_{network} = bottleneck_features['valid']
test_{network} = bottleneck_features['test']
In [25]:
### TODO: Obtain bottleneck features from another pre-trained CNN.
bottleneck_features = np.load('data/bottleneck_features/DogResnet50Data.npz')
train_resnet50 = bottleneck_features['train']
valid_resnet50 = bottleneck_features['valid']
test_resnet50 = bottleneck_features['test']

The above method to generate bottleneck features is not used in the project. When the bottleneck features were generated by Udacity, the bottleneck size for resnet50 used to be 1x1. This has been changed to 7x7 recently. Further details on the change are available here.

The bottleneck features for the training, validation and test sets are re-created below

In [26]:
from extract_bottleneck_features import *

train_resnet50 = extract_Resnet50(paths_to_tensor(train_files))
valid_resnet50 = extract_Resnet50(paths_to_tensor(valid_files))
test_resnet50 = extract_Resnet50(paths_to_tensor(test_files))
100%|█████████████████████████████████████████████████████████████████████████████| 6680/6680 [00:36<00:00, 184.59it/s]
C:\Users\UDAYG\.conda\envs\deep_learn\lib\site-packages\keras_applications\resnet50.py:265: UserWarning: The output shape of `ResNet50(include_top=False)` has been changed since Keras 2.2.0.
  warnings.warn('The output shape of `ResNet50(include_top=False)` '
100%|███████████████████████████████████████████████████████████████████████████████| 835/835 [00:04<00:00, 201.25it/s]
100%|███████████████████████████████████████████████████████████████████████████████| 836/836 [00:04<00:00, 200.85it/s]
In [27]:
print('training set shape:', train_resnet50.shape)
print('validation set shape: ', valid_resnet50.shape)
print('testing set shape: ', test_resnet50.shape)
training set shape: (6680, 7, 7, 2048)
validation set shape:  (835, 7, 7, 2048)
testing set shape:  (836, 7, 7, 2048)

(IMPLEMENTATION) Model Architecture

Create a CNN to classify dog breed. At the end of your code cell block, summarize the layers of your model by executing the line:

    <your model's name>.summary()

Question 5: Outline the steps you took to get to your final CNN architecture and your reasoning at each step. Describe why you think the architecture is suitable for the current problem.

Answer: I have decided to use the ResNet-50 pre-trained model.

To design the architecture, I have started with the architecture used for VGG16 above to see how we perform. Since this architecture uses a global average pooling layer followed by the output layer, it has the minimum number of parameters needed and so I wish to see the performance before adding / changing the layers. On adding more layers, I find that the validation loss increases, implying overfitting. As a result, I have decided to continue with a global average pooling layer followed by the output layer.

I have trained with 40 epochs with a checkpoint to early stop if the validation loss doesn't improve for 10 consecutive epochs. I have also plotted the training and validation loss against epochs, as well as, the training and validation accuracy against epochs to see when the model overfits. I see that the model starts to overfit after epoch 8. I get a testing accuracy of roughly 82% with the weights calculated in epoch 5 - these are the weights with the least validation loss. So I will continue with this architecture.

Even though, we use the minimum parameters to classify the bottleneck_features, this architecture still gives good results. I believe this is because Resnet-50 is a very deep network and as can be seen from the output shape of global_average_pooling_layer in the model summary, it produces bottleneck_features with a depth of 2048. This allows the network to identify the different features in the images needed for the dog breed classification.

The VGG16 based transfer training network above could not achieve this level of accuracy because it is not such a deep network and relies on fully connected layers with a reasonably high number of nodes to aid in the classification. In the Simonyan and Zisserman (2014) paper (see Source below), they had 2 hidden fully connected layers with 4096 nodes each, whereas, we have just a global average pooling layer.

The CNN in Step 3 is not deep enough to detect sufficient features needed for high accuracy and as a result, isn’t able to achieve high accuracy.

Source: https://www.pyimagesearch.com/2017/03/20/imagenet-vggnet-resnet-inception-xception-keras/

In [28]:
### TODO: Define your architecture.
resnet50_model = Sequential()
resnet50_model.add(GlobalAveragePooling2D(input_shape=train_resnet50.shape[1:]))
resnet50_model.add(Dense(133, activation='softmax'))

resnet50_model.summary()
Model: "sequential_3"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
global_average_pooling2d_2 ( (None, 2048)              0         
_________________________________________________________________
dense_4 (Dense)              (None, 133)               272517    
=================================================================
Total params: 272,517
Trainable params: 272,517
Non-trainable params: 0
_________________________________________________________________

(IMPLEMENTATION) Compile the Model

In [29]:
### TODO: Compile the model.
resnet50_model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])

(IMPLEMENTATION) Train the Model

Train your model in the code cell below. Use model checkpointing to save the model that attains the best validation loss.

You are welcome to augment the training data, but this is not a requirement.

In [30]:
### TODO: Train the model.
checkpointer_1 = ModelCheckpoint(filepath='saved_models/weights.best.Resnet50.hdf5', 
                               verbose=1, save_best_only=True)

checkpointer_2 = EarlyStopping(monitor = 'val_loss', verbose = 1, patience = 10)

calcs = resnet50_model.fit(train_resnet50, train_targets, 
          validation_data=(valid_resnet50, valid_targets),
          epochs=40, batch_size=20, callbacks=[checkpointer_1, checkpointer_2], verbose=1)

# summarize history for accuracy
plt.plot(calcs.history['accuracy'])
plt.plot(calcs.history['val_accuracy'])
plt.title('Model Accuracy')
plt.ylabel('Accuracy')
plt.xlabel('epoch')
plt.legend(['train', 'test'], loc='upper left')
plt.show()

# summarize history for loss
plt.plot(calcs.history['loss'])
plt.plot(calcs.history['val_loss'])
plt.title('Model Loss')
plt.ylabel('Loss')
plt.xlabel('epoch')
plt.legend(['train', 'test'], loc='upper left')
plt.show()
Train on 6680 samples, validate on 835 samples
Epoch 1/40
6680/6680 [==============================] - 2s 373us/step - loss: 1.6726 - accuracy: 0.5982 - val_loss: 0.7775 - val_accuracy: 0.7713

Epoch 00001: val_loss improved from inf to 0.77755, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 2/40
6680/6680 [==============================] - 2s 364us/step - loss: 0.3553 - accuracy: 0.9054 - val_loss: 0.6537 - val_accuracy: 0.7928

Epoch 00002: val_loss improved from 0.77755 to 0.65373, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 3/40
6680/6680 [==============================] - 2s 365us/step - loss: 0.1692 - accuracy: 0.9632 - val_loss: 0.6153 - val_accuracy: 0.8096

Epoch 00003: val_loss improved from 0.65373 to 0.61527, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 4/40
6680/6680 [==============================] - 2s 367us/step - loss: 0.0894 - accuracy: 0.9874 - val_loss: 0.5746 - val_accuracy: 0.8132

Epoch 00004: val_loss improved from 0.61527 to 0.57461, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 5/40
6680/6680 [==============================] - 2s 369us/step - loss: 0.0538 - accuracy: 0.9955 - val_loss: 0.5771 - val_accuracy: 0.8240

Epoch 00005: val_loss did not improve from 0.57461
Epoch 6/40
6680/6680 [==============================] - 2s 372us/step - loss: 0.0401 - accuracy: 0.9967 - val_loss: 0.5480 - val_accuracy: 0.8251

Epoch 00006: val_loss improved from 0.57461 to 0.54804, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 7/40
6680/6680 [==============================] - 2s 368us/step - loss: 0.0305 - accuracy: 0.9973 - val_loss: 0.5401 - val_accuracy: 0.8263

Epoch 00007: val_loss improved from 0.54804 to 0.54012, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 8/40
6680/6680 [==============================] - 2s 368us/step - loss: 0.0280 - accuracy: 0.9963 - val_loss: 0.5282 - val_accuracy: 0.8455

Epoch 00008: val_loss improved from 0.54012 to 0.52815, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 9/40
6680/6680 [==============================] - 2s 373us/step - loss: 0.0222 - accuracy: 0.9976 - val_loss: 0.5430 - val_accuracy: 0.8431

Epoch 00009: val_loss did not improve from 0.52815
Epoch 10/40
6680/6680 [==============================] - 2s 372us/step - loss: 0.0191 - accuracy: 0.9973 - val_loss: 0.5331 - val_accuracy: 0.8515

Epoch 00010: val_loss did not improve from 0.52815
Epoch 11/40
6680/6680 [==============================] - 2s 372us/step - loss: 0.0242 - accuracy: 0.9949 - val_loss: 0.5964 - val_accuracy: 0.8455

Epoch 00011: val_loss did not improve from 0.52815
Epoch 12/40
6680/6680 [==============================] - 3s 375us/step - loss: 0.0207 - accuracy: 0.9969 - val_loss: 0.5360 - val_accuracy: 0.8491

Epoch 00012: val_loss did not improve from 0.52815
Epoch 13/40
6680/6680 [==============================] - 2s 369us/step - loss: 0.0124 - accuracy: 0.9982 - val_loss: 0.5463 - val_accuracy: 0.8491

Epoch 00013: val_loss did not improve from 0.52815
Epoch 14/40
6680/6680 [==============================] - 2s 372us/step - loss: 0.0176 - accuracy: 0.9960 - val_loss: 0.6793 - val_accuracy: 0.8347

Epoch 00014: val_loss did not improve from 0.52815
Epoch 15/40
6680/6680 [==============================] - 2s 372us/step - loss: 0.0592 - accuracy: 0.9864 - val_loss: 0.8179 - val_accuracy: 0.8048

Epoch 00015: val_loss did not improve from 0.52815
Epoch 16/40
6680/6680 [==============================] - 3s 375us/step - loss: 0.0956 - accuracy: 0.9731 - val_loss: 1.0213 - val_accuracy: 0.7497

Epoch 00016: val_loss did not improve from 0.52815
Epoch 17/40
6680/6680 [==============================] - 2s 374us/step - loss: 0.0676 - accuracy: 0.9786 - val_loss: 0.8528 - val_accuracy: 0.8036

Epoch 00017: val_loss did not improve from 0.52815
Epoch 18/40
6680/6680 [==============================] - 2s 372us/step - loss: 0.0180 - accuracy: 0.9951 - val_loss: 0.7843 - val_accuracy: 0.8192

Epoch 00018: val_loss did not improve from 0.52815
Epoch 00018: early stopping

(IMPLEMENTATION) Load the Model with the Best Validation Loss

In [31]:
### TODO: Load the model weights with the best validation loss.
resnet50_model.load_weights('saved_models/weights.best.Resnet50.hdf5')

(IMPLEMENTATION) Test the Model

Try out your model on the test dataset of dog images. Ensure that your test accuracy is greater than 60%.

In [32]:
### TODO: Calculate classification accuracy on the test dataset.
# get index of predicted dog breed for each image in test set
resnet50_predictions = [np.argmax(resnet50_model.predict(np.expand_dims(feature, axis=0))) for feature in test_resnet50]

# report test accuracy
test_accuracy = 100*np.sum(np.array(resnet50_predictions)==np.argmax(test_targets, axis=1))/len(resnet50_predictions)
print('Test accuracy: %.4f%%' % test_accuracy)
Test accuracy: 82.0574%

(IMPLEMENTATION) Predict Dog Breed with the Model

Write a function that takes an image path as input and returns the dog breed (Affenpinscher, Afghan_hound, etc) that is predicted by your model.

Similar to the analogous function in Step 5, your function should have three steps:

  1. Extract the bottleneck features corresponding to the chosen CNN model.
  2. Supply the bottleneck features as input to the model to return the predicted vector. Note that the argmax of this prediction vector gives the index of the predicted dog breed.
  3. Use the dog_names array defined in Step 0 of this notebook to return the corresponding breed.

The functions to extract the bottleneck features can be found in extract_bottleneck_features.py, and they have been imported in an earlier code cell. To obtain the bottleneck features corresponding to your chosen CNN architecture, you need to use the function

extract_{network}

where {network}, in the above filename, should be one of VGG19, Resnet50, InceptionV3, or Xception.

In [33]:
### TODO: Write a function that takes a path to an image as input
### and returns the dog breed that is predicted by the model.

from extract_bottleneck_features import *

def Resnet50_predict_breed(img_path):
    # extract bottleneck features
    bottleneck_feature = extract_Resnet50(path_to_tensor(img_path))
    # obtain predicted vector
    predicted_vector = resnet50_model.predict(bottleneck_feature)   
    # return dog breed that is predicted by the model
    return dog_names[np.argmax(predicted_vector[0])]

Step 6: Write your Algorithm

Write an algorithm that accepts a file path to an image and first determines whether the image contains a human, dog, or neither. Then,

  • if a dog is detected in the image, return the predicted breed.
  • if a human is detected in the image, return the resembling dog breed.
  • if neither is detected in the image, provide output that indicates an error.

You are welcome to write your own functions for detecting humans and dogs in images, but feel free to use the face_detector and dog_detector functions developed above. You are required to use your CNN from Step 5 to predict dog breed.

Some sample output for our algorithm is provided below, but feel free to design your own user experience!

Sample Human Output

(IMPLEMENTATION) Write your Algorithm

In [34]:
### TODO: Write your algorithm.
### Feel free to use as many code cells as needed.

import matplotlib.pyplot as plt
%matplotlib inline

def show_image(img_path):
    img = plt.imread(img_path)
    plt.imshow(img)
    plt.show()
    return

def dog_name(img_path):
    if dog_detector(img_path):
        dog_breed = Resnet50_predict_breed(img_path)
        print('Hello Dog')
        show_image(img_path)
        print('Your predicted breed is ', dog_breed.split('.')[1])
    elif face_detector(img_path):
        dog_resemble = Resnet50_predict_breed(img_path)
        print('Hello Human')
        show_image(img_path)
        print('You look like a...\n', dog_resemble.split('.')[1])
    else:
        print('There is an error. You look neither like a dog nor a human')
        show_image(img_path)
        return

Step 7: Test Your Algorithm

In this section, you will take your new algorithm for a spin! What kind of dog does the algorithm think that you look like? If you have a dog, does it predict your dog's breed accurately? If you have a cat, does it mistakenly think that your cat is a dog?

(IMPLEMENTATION) Test Your Algorithm on Sample Images!

Test your algorithm at least six images on your computer. Feel free to use any images you like. Use at least two human and two dog images.

Question 6: Is the output better than you expected :) ? Or worse :( ? Provide at least three possible points of improvement for your algorithm.

Answer: I am surprised by the results as we do broadly have a good classification of dog vs human vs others. The dog breed classification is not great as we have many errors. There are few areas where there can be improvements:

  1. Some classifications are not correct. For example, the Chinese Shar Pei is classified as Neapolitan Mastiff, English Setter is classified as Clumber Spaniel. There are two improvements we can have:

    A) The fully connected layers could be improved for better classification.

    B) We can also include data augmentation.

  2. The human face detection algorithm didn't work for a picture of Thor with helmet. A reason could be that the helmet covers a lot of features such as ears, hair, general shape of face, which results in a negative result. Potentially the algorithm could be improved by training with some caps or helmet type features.

  3. The human face detection algorithm also detects arctic fox (and another fox) as a human. This could be improved either through a better algorithm or potentially creating a better trained neural network.
In [35]:
## TODO: Execute your algorithm from Step 6 on
## at least 6 images on your computer.
## Feel free to use as many code cells as needed.
import os

step7_dir = os.path.join(os.getcwd(), 'Step7 Images')
files = [f for f in os.listdir(step7_dir) if os.path.isfile(os.path.join(step7_dir, f))]
for f in files:
    img_path = 'Step7 Images/'+f
    print('Filename: ', f)
    dog_name(img_path)
    print('\n')
Filename:  Akita.jpg
C:\Users\UDAYG\.conda\envs\deep_learn\lib\site-packages\keras_applications\resnet50.py:265: UserWarning: The output shape of `ResNet50(include_top=False)` has been changed since Keras 2.2.0.
  warnings.warn('The output shape of `ResNet50(include_top=False)` '
Hello Dog
Your predicted breed is  Akita


Filename:  american water spaniel.jpeg
Hello Dog
Your predicted breed is  American_water_spaniel


Filename:  Arctic Fox  Alopex lagopus.jpg
Hello Human
You look like a...
 American_eskimo_dog


Filename:  beagle.jpg
Hello Dog
Your predicted breed is  Beagle


Filename:  belgium_shepherd_dog.jpg
Hello Dog
Your predicted breed is  Belgian_sheepdog


Filename:  bernese-mountain-dog.jpg
Hello Dog
Your predicted breed is  Bernese_mountain_dog


Filename:  cat 2.jpg
There is an error. You look neither like a dog nor a human

Filename:  cat.jpeg
There is an error. You look neither like a dog nor a human

Filename:  chihuahua_smooth_coat.jpg
Hello Dog
Your predicted breed is  Chihuahua


Filename:  curly coated retriever.png
Hello Dog
Your predicted breed is  Curly-coated_retriever


Filename:  Dachshund.jpg
Hello Dog
Your predicted breed is  Dachshund


Filename:  Dalmatian.jpg
Hello Dog
Your predicted breed is  Dalmatian


Filename:  Dog-Chinese_Shar_Pei.jpg
Hello Dog
Your predicted breed is  Neapolitan_mastiff


Filename:  donald trump.jpg
Hello Human
You look like a...
 Silky_terrier


Filename:  English-Setter.jpg
Hello Dog
Your predicted breed is  Clumber_spaniel


Filename:  Finnish-Spitz.jpg
Hello Dog
Your predicted breed is  Finnish_spitz


Filename:  fox-laurie-campbell.jpg
Hello Human
You look like a...
 Norwegian_lundehund


Filename:  german_shepherd.jpg
Hello Dog
Your predicted breed is  German_shepherd_dog


Filename:  Great-Pyrenees.jpg
Hello Dog
Your predicted breed is  Great_pyrenees


Filename:  great_dane.jpg
Hello Dog
Your predicted breed is  Great_dane


Filename:  HOUND_Afghan-Hound.jpg
Hello Dog
Your predicted breed is  Afghan_hound


Filename:  irish-wolfhound.jpg
Hello Dog
Your predicted breed is  Irish_wolfhound


Filename:  japanese-chin.jpg
Hello Dog
Your predicted breed is  Japanese_chin


Filename:  leonberger.jpg
Hello Dog
Your predicted breed is  Leonberger


Filename:  manchester terrier.jpg
Hello Dog
Your predicted breed is  Manchester_terrier


Filename:  myself.png
Hello Human
You look like a...
 Bullmastiff


Filename:  norwegian-lundehund-1.jpg
There is an error. You look neither like a dog nor a human

Filename:  portuguese water dog.jpg
Hello Dog
Your predicted breed is  Black_russian_terrier


Filename:  rabbit.png
There is an error. You look neither like a dog nor a human

Filename:  raccoon_rocket.jpg
There is an error. You look neither like a dog nor a human

Filename:  thor.jpg
There is an error. You look neither like a dog nor a human

Filename:  Tibetan-Mastiff-On-White-03.jpg
Hello Dog
Your predicted breed is  Tibetan_mastiff


Filename:  tiger.jpeg
There is an error. You look neither like a dog nor a human

Filename:  Welsh-Springer-Spaniel.jpg
Hello Dog
Your predicted breed is  Welsh_springer_spaniel


Filename:  wirehaired-pointing-griffon.jpg
Hello Dog
Your predicted breed is  Irish_wolfhound


Filename:  Wolf.jpg
There is an error. You look neither like a dog nor a human

In [36]:
# Dog Breed Names included in the Training Set
dog_names
Out[36]:
['n\\001.Affenpinscher',
 'n\\002.Afghan_hound',
 'n\\003.Airedale_terrier',
 'n\\004.Akita',
 'n\\005.Alaskan_malamute',
 'n\\006.American_eskimo_dog',
 'n\\007.American_foxhound',
 'n\\008.American_staffordshire_terrier',
 'n\\009.American_water_spaniel',
 'n\\010.Anatolian_shepherd_dog',
 'n\\011.Australian_cattle_dog',
 'n\\012.Australian_shepherd',
 'n\\013.Australian_terrier',
 'n\\014.Basenji',
 'n\\015.Basset_hound',
 'n\\016.Beagle',
 'n\\017.Bearded_collie',
 'n\\018.Beauceron',
 'n\\019.Bedlington_terrier',
 'n\\020.Belgian_malinois',
 'n\\021.Belgian_sheepdog',
 'n\\022.Belgian_tervuren',
 'n\\023.Bernese_mountain_dog',
 'n\\024.Bichon_frise',
 'n\\025.Black_and_tan_coonhound',
 'n\\026.Black_russian_terrier',
 'n\\027.Bloodhound',
 'n\\028.Bluetick_coonhound',
 'n\\029.Border_collie',
 'n\\030.Border_terrier',
 'n\\031.Borzoi',
 'n\\032.Boston_terrier',
 'n\\033.Bouvier_des_flandres',
 'n\\034.Boxer',
 'n\\035.Boykin_spaniel',
 'n\\036.Briard',
 'n\\037.Brittany',
 'n\\038.Brussels_griffon',
 'n\\039.Bull_terrier',
 'n\\040.Bulldog',
 'n\\041.Bullmastiff',
 'n\\042.Cairn_terrier',
 'n\\043.Canaan_dog',
 'n\\044.Cane_corso',
 'n\\045.Cardigan_welsh_corgi',
 'n\\046.Cavalier_king_charles_spaniel',
 'n\\047.Chesapeake_bay_retriever',
 'n\\048.Chihuahua',
 'n\\049.Chinese_crested',
 'n\\050.Chinese_shar-pei',
 'n\\051.Chow_chow',
 'n\\052.Clumber_spaniel',
 'n\\053.Cocker_spaniel',
 'n\\054.Collie',
 'n\\055.Curly-coated_retriever',
 'n\\056.Dachshund',
 'n\\057.Dalmatian',
 'n\\058.Dandie_dinmont_terrier',
 'n\\059.Doberman_pinscher',
 'n\\060.Dogue_de_bordeaux',
 'n\\061.English_cocker_spaniel',
 'n\\062.English_setter',
 'n\\063.English_springer_spaniel',
 'n\\064.English_toy_spaniel',
 'n\\065.Entlebucher_mountain_dog',
 'n\\066.Field_spaniel',
 'n\\067.Finnish_spitz',
 'n\\068.Flat-coated_retriever',
 'n\\069.French_bulldog',
 'n\\070.German_pinscher',
 'n\\071.German_shepherd_dog',
 'n\\072.German_shorthaired_pointer',
 'n\\073.German_wirehaired_pointer',
 'n\\074.Giant_schnauzer',
 'n\\075.Glen_of_imaal_terrier',
 'n\\076.Golden_retriever',
 'n\\077.Gordon_setter',
 'n\\078.Great_dane',
 'n\\079.Great_pyrenees',
 'n\\080.Greater_swiss_mountain_dog',
 'n\\081.Greyhound',
 'n\\082.Havanese',
 'n\\083.Ibizan_hound',
 'n\\084.Icelandic_sheepdog',
 'n\\085.Irish_red_and_white_setter',
 'n\\086.Irish_setter',
 'n\\087.Irish_terrier',
 'n\\088.Irish_water_spaniel',
 'n\\089.Irish_wolfhound',
 'n\\090.Italian_greyhound',
 'n\\091.Japanese_chin',
 'n\\092.Keeshond',
 'n\\093.Kerry_blue_terrier',
 'n\\094.Komondor',
 'n\\095.Kuvasz',
 'n\\096.Labrador_retriever',
 'n\\097.Lakeland_terrier',
 'n\\098.Leonberger',
 'n\\099.Lhasa_apso',
 'n\\100.Lowchen',
 'n\\101.Maltese',
 'n\\102.Manchester_terrier',
 'n\\103.Mastiff',
 'n\\104.Miniature_schnauzer',
 'n\\105.Neapolitan_mastiff',
 'n\\106.Newfoundland',
 'n\\107.Norfolk_terrier',
 'n\\108.Norwegian_buhund',
 'n\\109.Norwegian_elkhound',
 'n\\110.Norwegian_lundehund',
 'n\\111.Norwich_terrier',
 'n\\112.Nova_scotia_duck_tolling_retriever',
 'n\\113.Old_english_sheepdog',
 'n\\114.Otterhound',
 'n\\115.Papillon',
 'n\\116.Parson_russell_terrier',
 'n\\117.Pekingese',
 'n\\118.Pembroke_welsh_corgi',
 'n\\119.Petit_basset_griffon_vendeen',
 'n\\120.Pharaoh_hound',
 'n\\121.Plott',
 'n\\122.Pointer',
 'n\\123.Pomeranian',
 'n\\124.Poodle',
 'n\\125.Portuguese_water_dog',
 'n\\126.Saint_bernard',
 'n\\127.Silky_terrier',
 'n\\128.Smooth_fox_terrier',
 'n\\129.Tibetan_mastiff',
 'n\\130.Welsh_springer_spaniel',
 'n\\131.Wirehaired_pointing_griffon',
 'n\\132.Xoloitzcuintli',
 'n\\133.Yorkshire_terrier']